BACKGROUND: Neonatal illness, particularly in the first week of life, is a leading cause of death worldwide. Improving identification of young infants who require referral for severe illness is of major public-health importance. METHODS: Infants under 2 months of age brought with illness to health facilities in Bangladesh, Bolivia, Ghana, India, Pakistan, and South Africa were recruited in two age-groups: 0-6 days and 7-59 days. A trained health worker recorded 31 symptoms and clinical signs. An expert paediatrician assessed each case independently for severe illness that required hospital admission. We examined the sensitivity, specificity, and odds ratio (OR) for each symptom and sign individually and combined into algorithms to assess their value for predicting severe illness, excluding jaundice. FINDINGS: 3177 children aged 0-6 days and 5712 infants aged 7-59 days were enrolled. 12 symptoms or signs predicted severe illness in the first week of life: history of difficulty feeding (OR 10.0, 95% CI, 6.9-14.5), history of convulsions (15.4, 6.4-37.2), lethargy (3.5, 1.7-7.1), movement only when stimulated (6.9, 3.0-15.5), respiratory rate of 60 breaths per minute or more (2.7, 1.9-3.8), grunting (2.9, 1.1-7.5), severe chest indrawing (8.9, 4.0-20.1), temperature of 37.5 degrees C or more (3.4, 2.4-4.9) or below 35.5 degrees C (9.2, 4.6-18.6), prolonged capillary refill (10.5, 5.1-21.7), cyanosis (13.7, 1.6-116.5), and stiff limbs (15.1, 2.2-105.9). A decision rule requiring the presence of any one sign had high sensitivity (87%) and specificity (74%). After we reduced the algorithm to seven signs (history of difficulty feeding, history of convulsions, movement only when stimulated, respiratory rate of 60 breaths per minute or more, severe chest indrawing, temperature of 37.5 degrees C or more or below 35.5 degrees C), mainly on the basis of prevalence of each sign or symptom, sensitivity (85%) and specificity (75%) were much the same. These seven signs also did well in 7-59-day-old infants (sensitivity 74%, specificity 79%). INTERPRETATION: A single simple algorithm could be recommended for identifying severe illness in infants aged 0-2 months who are brought to health facilities. Further research is needed on screening newborn children for illness in the community during routine home visits.